Individualized prevention of proton pump inhibitor related adverse events by risk stratification

Abstract

Proton pump inhibitors (PPIs) are commonly used for gastric acid-related disorders, but their safety profile and risk stratification for high-burden diseases need further investigation. Analyzing over 2 million participants from five prospective cohorts from the US, the UK, and China, we found that PPI use correlated with increased risk of 15 leading global diseases, such as ischemic heart disease, diabetes, respiratory infections, and chronic kidney disease. These associations showed dose-response relationships and consistency across different PPI types. PPI-related absolute risks increased with baseline risks, with approximately 82% of cases occurring in those at the upper 40% of the baseline predicted risk, and only 11.5% of cases occurring in individuals at the lower 50% of the baseline risk. While statistical association does not necessarily imply causation, its potential safety concerns suggest that personalized use of PPIs through risk stratification might guide appropriate decision-making for patients, clinicians, and the public.

Fig. 1. Flowchart of participant inclusion.

This prospective analysis encompasses five population-based cohort studies: Clinical Data Analysis and Reporting System (CDARS), China; Nurses’ Health Study (NHS), USA; Nurses’ Health StudyII (NHS II), USA;Health Professionals Follow-Up Study (HPFS), USA; and UK Biobank, UK. We extracted data from 4,903,903 individuals from the CDARS who had information on the prescription of either a Proton pump inhibitor (PPI) or H2RA from January 1, 2003, through December 31, 2017. Exclusion criteria included individuals classified as short-term or non-users of PPI, those who used PPI within two years prior to cohort entry, and instances of data linkage errors. After exclusions, the CDARS baseline cohort consisted of 1,356,333 individuals. Similar exclusions (e.g., lack of follow-up or other reasons) were applied to the other cohorts, resulting in 91,708 individuals for NHS, 99,641 individuals for NHS II, 30,933 individuals for HPFS, and 501,109 individuals for UK Biobank. The final baseline study population included 2,079,724 participants across all five studies. For analysis of 30 leading causes of global disease burden, this study included cohorts with corresponding disease data, with the exclusion of participants with prevalent diseases or related phenotypes prior to baseline and availability of follow-up information on the date of event diagnosis for subsequent analysis.

Fig. 2. Combined analyses of PPI use and risk of the 30 leading causes of global disease burden.

The squares denote the adjusted hazard ratio (HR), and the horizontal lines represent the 95% confidence intervals (CI). Two-sided P values were derived using inverse variance-weighted, random-effect meta-analyses (N = 2,079,724) and are not corrected for multiple testing. U UK Biobank, N NHS, N2 NHS II, H HPFS, C CDARS.

Fig. 3. Dose–response associations according to the accumulated duration of PPI use in CDARS.

The squares denote the adjusted hazard ratio (HR), and the vertical lines represent the 95% confidence intervals (CI). All displayed P values are two-sided for trend analyses (N = 1,356,333) without adjustment for multiple comparisons. Source data are provided as a Source Data file.

Fig. 4. Risk stratification for PPI-related adverse events.

The analysis was carried out for the 15 high-burden diseases that were associated with PPI use in the primary analysis based on the UK Biobank (N = 501,109). The baseline risk was evaluated with newly established models (lower respiratory infections, falls, diarrheal diseases, cirrhosis and other chronic liver diseases, asthma) or previously reported prediction models (Supplementary Data 1). We evaluated the HRs of PPI use as compared with no-PPI use in each quartile group, then translated to absolute effects (risk differences, RDs) associated with PPI use at one year by the method described by Altman. The squares denote the RD, and the vertical lines represent the 95% confidence intervals (CI). Source data are provided as a Source Data file.

Fig. 5. Absolute risk of PPI-related high-burden diseases according to the distribution of the baseline predicted risk.

Abbreviation: AF, attributable fraction; PPI, proton pump inhibitor; RD, risk difference. This histogram presented the distribution of baseline predicted risk for any of the 15 PPI-related diseases. The performance and nomogram for the prediction model is available in Supplementary Table S36 and Supplementary Figs. S7 and 8. The RD and AF of PPI use for 1 year was calculated. Annual number of cases attributed to PPI use in each strata (i.e., n) was calculated based on attributable risk and exposure, and then summed as the total annual number of cases attributed to PPI use in all populations (i.e., total). The results showed that most cases were occurred in the individuals with high baseline predicted risk, and those who with low baseline risk do not need to be over panicked and should adhere to PPI treatment. Source data are provided as a Source Data file.